Anthropic accuses Alibaba of 29M-query AI model theft campaign

Anthropic alleges massive extraction campaign

AI developer Anthropic has accused Chinese technology giant Alibaba of orchestrating what it calls the largest campaign to illicitly extract its Claude AI model's capabilities. In a June 10 letter to US Senators Tim Scott and Elizabeth Warren, the San Francisco company detailed nearly 29 million exchanges with Claude conducted through thousands of fraudulent accounts.

The campaign employed "distillation attacks," a technique where operators query a sophisticated AI model repeatedly to capture its responses, then use those answers to train a less capable model. Anthropic said the attacks specifically targeted Claude's most valuable features, including its ability to handle complex, extended tasks and its decision-making frameworks.

According to the letter, operators linked to the New York Stock Exchange-listed Alibaba conducted these extractions on an "industrial scale" to enable Chinese companies to harvest and rebrand US AI capabilities as their own products.

Why it matters

Distillation attacks represent a direct threat to the competitive advantage of US AI companies that have invested hundreds of billions in research and development. If foreign competitors can replicate advanced AI capabilities at a fraction of the original cost, it undermines the business model of American AI leaders and potentially shifts technological leadership overseas. The accusation also escalates tensions between US and Chinese tech sectors at a moment when both nations are racing for AI dominance.

Broader national security concerns

Anthropic's letter went beyond the Alibaba allegations to cite what it described as threats to US military interests. The company referenced Department of Defense claims that Alibaba and other major Chinese firms—including automaker BYD and tech company Baidu—maintain ties to the Chinese military. Those companies have denied such allegations.

Alibaba is currently suing the US government to remove its name from the Pentagon's blacklist of companies with alleged military connections.

Anthropic urged Congress to penalize companies behind such attacks and strengthen measures to prevent theft of US technology. The company argued that distillation attacks effectively convert American investment into "a massive subsidy for our geopolitical competitors."

Pattern of alleged extraction

US AI developers have previously raised similar concerns about Chinese competitors using distillation techniques to train rival models. OpenAI, maker of ChatGPT, has also accused Chinese groups of employing the same practice to replicate American AI technology at reduced cost.

Anthropic, a leading AI developer alongside OpenAI, is preparing for a stock market debut that analysts expect could value it among the world's most valuable companies. However, some of the company's advanced models, including one called Mythos, have raised cybersecurity questions about their ability to identify vulnerabilities in computer systems.

The BBC reported these details and has contacted both Alibaba and Anthropic for additional comment.